A lot is being made about AI and LLMs. There is good reason for this.
As we dive deeper into Web 3.0, we have to realize what is means to create a decentralized Internet. For those who are focused upon LEO, this means LeoAI.
We see the Internet having AI added all over the place. While it is probably a bit pre-mature in some instances, we can see how Google, Meta, and X are incorporating it into their platforms.
This means that Web 3.0 must do the same. After all, if the tools are getting better on Web 2.0, why would anyone switch?
However, there are some present drawbacks to these models. For that reason, we will dive into what we are dealing with.
https://inleo.io/threads/view/taskmaster4450le/re-lfsrsdssef
LeoAI: Data To Context
We talk a great deal about getting data into a permissionless database. This is really just the beginning.
The above short go through the 3 phases of data. As we can see they are:
- data
- information
- knowledge
The last one is where we want to focus our attention.
Before getting to that, lets briefly run through the other two.
Data is raw. This is simply what is entered into a database. While there is could be some structure to it, such as with tables, for the most part it is a mess. This is especially true for data from social media platforms.
Information is the second tier. This is where data is cleaned up and made useful. Labelling is done so the model grasps what it is dealing with. Vector databases are created, connecting the information based upon it relevancy to the other sections.
The final area is knowledge. This can be summed up with the word content. All the data in the world means nothing if it has no context. This is where many models are presently struggling.
Of course, as time elapses, this changes. More information actually helps to form conceptions of what we are dealing with. This is why adding more data is crucial. As more topics are discussed, it becomes a framework from which to formulate understanding.
This is basically what humans do.
Image generated by Ideogram
Feed The Beast
When it comes to data, AI is a hungry animal.
Many are speculating that we are going to run out of data in the next couple years. Each iteration of models requires more than was previously used. Personally, when embedded AI starts to roll out (mostly through robots), I fail to see how this will be a problem.
That said, we need to keep feeding the beast simply to help it have context. Raw data does not amount to much. Even in the two dimensional realm, there are limits. However, unless one has a robot to use, this is what we are stuck with.
It does not mean all is lost. Actually, to the contrary. These models are improving, meaning it is better able to understand what is fed it. However, it still requires massive amount of human generated content to help to conceptualize what us humans deal with. After all, most of what we have was designed by humans.
For example, how would AI understand the corporation? What is meant by it? We could symbolize it with a large building in a major city. Or a logo of one of the more well known companies could be used. Naturally, a corporation is really just articles on a piece of paper (computer).
As you can see, a lot is required to even have the slightest understanding of what we are dealing with.
This is why feeding in data is crucial. We are not training it on the corporation in particular. Instead, we have to help to comprehend the physical from the real and compare it to the digital. A corporation really doesn't exist in the physical world (a building or plant does though) yet it is part of the real world.
Here we are dealing with just one example.
If we want LeoAI to have strong utility, it is important to give it context. For example, what is Hive? Compare this to what a bee might experience. As we know, there are differences.
The more that is fed into the engine, more information is generated. From this, we hope to help foster a path towards context, i.e. knowledge, about the information it receives.
This is also why it is crucial to fill the database with information in different languages. Simply translating something will often miss the context and nuances of the language.
It is crucial the model be trained on this also.
Posted Using InLeo Alpha